AI in healthcare: adoption statistics 2026

Key findings

1. 1

   72% of US physicians actively use at least one AI tool in 2026, up from 38% in 2023. Counting those still evaluating it, 81% (AMA).

2. 2

   Adoption rises with size: 96% of large hospitals (over 400 beds) use predictive AI versus 59% of small ones, and 86% of system-affiliated hospitals versus 37% of independents (ONC, 2024).

3. 3

   AI use is documentation-first: billing and note documentation (21%) and discharge or care-plan notes (20%) lead assistive diagnosis (12%) among adopting physicians (AMA, 2024).

4. 4

   Ambient AI scribes saved Kaiser Permanente an estimated 15,700 documentation hours across 7,260 physicians and 2.5M encounters in one year. A separate 263-clinician study cut after-hours charting 0.9 hours a day and lowered burnout odds sharply.

5. 5

   The only randomized trial found smaller, product-dependent effects (a 9.5% note-time cut for one scribe, no significant change for another), so treat "saves an hour a day" as a marketing claim, not evidence.

6. 6

   Physicians lead with trust, not cost: 88% want a channel to flag AI problems, 85% want data-privacy assurance, 82% want independent validation. ROI ranks last at 68% (AMA, 2024).

7. 7

   Health-system AI is governance-gated: only 16% have a system-wide AI policy, and 40% of staff have encountered an unsanctioned "shadow AI" tool (CCM/KLAS and Wolters Kluwer, 2025).

8. 8

   The FDA has authorized about 1,451 AI-enabled medical devices, roughly 76% of them radiology, as of end-2025.

9. 9

   Patients accept AI when a human stays in charge: clinician presence made 3,000 US adults 18.4% more likely to choose an AI-involved visit, more than FDA approval or any certification (JAMA Network Open, 2026).

How many US physicians use AI in 2026?

72% actively use at least one AI tool, up from 38% in 2023. Counting those still evaluating it, 81%.

Adoption roughly doubled in three years, and the growth is documentation tools clearing the trust bar rather than diagnostic AI. Note the honest number: the 81% headline blends active use with physicians who are only evaluating AI, so the figure that survives scrutiny is 72% actually running a use case. There was no 2025 survey wave, so the middle of the curve is the source own interpolation, not a fresh measurement.

Who is adopting, hospitals or small practices?

The big systems. 96% of large hospitals use predictive AI versus 59% of small ones, and 86% of system-affiliated hospitals versus 37% of independents.

The gap is budget and integration capacity, not attitude. A large system can absorb a multi-month EHR integration and a governance committee. A small or independent hospital cannot, and ONC calls the result a persistent digital divide. Representative adoption data for solo and small ambulatory practices barely exists, which is itself the story: the smallest providers are the least measured and the least served.

What is AI actually used for in clinics?

Paperwork. Documentation (21%) and discharge or care-plan notes (20%) lead assistive diagnosis (12%).

Clinical AI gets the headlines, administrative AI gets the contracts. Documentation leads because its return is measured in minutes per day, not in diagnostic-accuracy debates that invite liability. The first AI project that survives procurement is almost always paperwork. These are confirmed 2024 figures. A richer 2026 ranking circulated but did not survive verification, so it is deliberately left off this page.

What do clinicians want before they adopt?

Trust, not price. 88% want a way to flag AI problems, 85% want data-privacy assurance, 82% want independent validation. ROI ranks last, at 68%.

This reorders the whole sales conversation. Physicians do not lead with cost, they lead with a governance and safety story: a feedback channel, a privacy guarantee, independent validation, liability cover. ROI is dead last. On the health-system side the same instinct shows up as immaturity rather than demand: only 16% have a system-wide AI policy and 40% of staff have already hit an unsanctioned shadow-AI tool. The vendor that ships a BAA, an audit trail and a validation report answers the top three objections before price ever comes up.

Will patients accept AI?

Yes, when a human stays in charge. Clinician oversight raised willingness to accept an AI visit by 18.4 points, more than FDA approval or any certification.

Patients are not against medical AI, they are against unsupervised medical AI. In a 3,000-person experiment, keeping a clinician in the loop moved acceptance more than an FDA clearance or a Mayo Clinic stamp. For anyone building healthcare AI, that is the product spec: the human-in-the-loop is not a compliance afterthought, it is the single biggest driver of whether patients will say yes.

Two numbers for context

Methodology

This page aggregates primary sources into one reference: the AMA Augmented Intelligence physician survey series (2023-2026 waves), the ONC/ASTP hospital predictive-AI data brief, the FDA AI-enabled device list, peer-reviewed ambient-scribe studies (NEJM Catalyst and JAMA Network Open), and Rock Health digital-health funding data. Every figure carries its source, sample size and field date. Where the only randomized evidence contradicts real-world deployment claims, both are shown. Projections and funding figures are labeled as such and never presented as measured market size. Adoption survey waves are repeated cross-sections, not one panel, so year-over-year comparisons follow the sources own framing. Reviewed quarterly.